Apple just made your AI vendor question routable
Apple put OpenAI, Anthropic, and Google behind Siri AI on Monday. Microsoft built the opposite stack at Build last week.
At WWDC on Monday, Apple put three AI model providers behind a single Siri surface and made the choice between them a developer-configurable variable. The same week, Microsoft shipped its own model on its own chip inside its own distribution.
In today's issue:
Main story: Apple just made your AI vendor question routable
Since Friday: Anthropic files for IPO, SpaceX signs an $11B annual compute deal with Google, and Yoshua Bengio calls for a coordinated pause on recursive self-improvement
At WWDC on Monday morning, Apple unveiled Siri AI and confirmed that developers building on iOS 27 can call Google's AI models alongside OpenAI's and Anthropic's, plus Apple's own on-device models. Google was the new name on the list, and the product ships in September on a developer surface that is live today.
The keynote was Tim Cook's last as CEO. He steps down September 1, and hardware veteran John Ternus takes over the same week iOS 27 ships. The product Cook spent ten years failing to get right ships under the next CEO, on a platform that does not depend on Apple winning any single AI model race.
Setup
The Siri rollout is a real product story, with a dedicated app and conversation sync across devices through iCloud. The activation switched from "Hey Siri" to gaze plus speech, the same friction reduction ChatGPT and Gemini have on phones, and visual intelligence is built on Apple's on-device models. The first reactions split sharply: some attendees called it the most important WWDC in over a decade, and others said the 2026 features should have shipped in 2025 and that Siri AI still trails Gemini on raw capability.
That argument is the wrong one to track. The operating event is that Apple shipped a consumer AI product that does not bet on any single model winning. Three named providers sit behind the same Siri surface, developers pick the one that fits the workload, and Apple takes the relationship with the consumer while leaving the model layer routable.
The turn
The standard read of WWDC is "Apple finally caught up." The read that matters is that the second-largest software company in the world just took the opposite position from the largest one, on whether AI is vertical or routable, inside the same eight-day window.
Last Tuesday at Build, Microsoft put its own model on its own chip in its own runtime inside its own distribution. Yesterday at WWDC, Apple put three model providers behind a single consumer surface and treated none of them as the answer. The question every operator now has to answer is which posture the company is buying into, on which workload, and whether the contracts already signed allow either to change.
The multi-provider stack is the product, not the model.
Apple's announcement names the providers and the access mechanism. App Intent through Siri AI lets a developer create events, hand images in as input, build custom skills, and choose which model handles which call. Privacy needs go to on-device, reasoning depth goes to the frontier provider that fits, and the procurement decision is no longer which lab the company believes in, but which routing layer the company controls.
A year ago, the consumer AI conversation was "Apple is two years behind OpenAI." The same conversation now has Apple shipping a product that turns the choice between OpenAI, Anthropic, and Google into a developer-configurable variable. The buyer side of that variable is the operator who runs the company that wrote the app.
The cost story has already moved.
The same Monday that Apple announced its multi-provider Siri, Tomasz Tunguz published a working list of named companies that have switched off frontier closed models in production. Cursor post-trained Kimi K2.5 into its own Composer 2.5 and reports a 10x efficiency gain. Harvey reports beating Opus on legal benchmarks at 11x lower cost. Lindy moved 100 percent of traffic to DeepSeek v4 and reports saving millions. Coinbase made a similar switch. OpenRouter named June "Cost Reduction Month" because the pattern always follows a wave of strong open-source releases.
The point is not that open-source has won. It is these named enterprise buyers who are choosing models the way a logistics operator chooses carriers, on price, on reliability on the specific workload, and on the cost of switching. Apple shipped a consumer surface that assumes this is how the market works, and Microsoft shipped a stack that assumes it is not.
Microsoft and Apple now disagree about the same question.
Microsoft's Build keynote a week ago landed seven first-party MAI releases, with MAI-Thinking-1 trained from scratch on Microsoft's own data and running on Microsoft's own Maia 200 silicon. The CIO question Microsoft is answering is "what if your AI provider raises prices, changes terms, or gets acquired?" The Microsoft answer is "we own the layer underneath the provider."
The Apple answer is the opposite. "The provider layer is going to be cheaper and more contested every quarter, so build a surface that lets you pick the cheapest acceptable model on the workload that matters." Neither company is obviously wrong. The 50-to-500-person operator does not get to pick which company is right at the platform level. The operator only gets to pick which posture the company's own AI contracts inherit, and whether the contracts allow the posture to change inside a renewal window.
AI READY PRO · FREE
A 30-day AI program. In your inbox. Then it ends.
One email every morning at 7 am ET. Each one is a short read and one real thing to try before lunch. By Day 30, you have nine concrete capabilities, including prompting that hits your bar on the first draft, AI workflows that take five hours a week off your plate, and the language to lead the AI conversation at your company.
Built for ops leaders, COOs, chiefs of staff, founders, and team leads at mid-market companies. The four weeks move you from foundations (effective prompting, catching hallucinations, learning AI with AI) to connect (a personal context layer, daily automation, five hours a week back), to build (reusable skills, agent design and debugging), to lead (tool and model evaluation, leading the conversation at work). Twenty-one mornings, no upsells, no community Slack.
Counterargument
The strongest objection is that this is overreading a developer surface. Apple has done multi-provider integrations before. ChatGPT through Siri was already in iOS 26. Adding Google and Anthropic to the developer API is a feature update, not a strategic re-orientation. The Microsoft contrast is real on paper and irrelevant on the ground, because enterprise contracts move on 24-month cycles, and most operators will not switch posture inside this fiscal year.
I want to be honest about the bear case. The counterargument is right that posture does not change in a quarter, and it is wrong that the contracts do not change. Three providers behind one surface puts a usable second source on the table for every operator buying enterprise AI, and the moment a usable second source exists, every renewal that comes up the rest of the year gets repriced. The Tunguz list is the leading indicator, and the Apple decision tips operators who were waiting for one of the big platforms to take the routable side.
What to do this week
Pull the next three AI contracts on the renewal calendar and answer one question for each. If your team needed to swap the underlying model provider in 90 days, what would have to be true. The answer "we would rewrite the integration" means the contract is locked in a way that the contract itself does not say. "We would renegotiate with one vendor" means the contract is routable, but the routing layer is theirs. "We would change a config and re-run a regression suite" means the contract is routable and the routing layer is yours. The third answer is the one that lets the company act on the cost moves the Tunguz list is showing.
Then have one conversation with whoever owns the AI vendor relationship. Ask whether the contract on renewal protects the company against a 50 percent price reduction in a comparable model. Most current AI contracts do not. The fix is to put the next renewal on a 6 or 12-month term, not 24, and add a competitive-model clause. Procurement teams know how to write that clause for SaaS, and they have not been writing it for AI.
The Build and WWDC keynotes sit a week apart for a reason. The operator's move is not to pick which company wins the platform argument; it is to make sure the next contract you sign survives either outcome.
From the field
The pattern I keep seeing in client engagements over the last quarter is that the AI contract in the pilot binder was written for the AI in 2024, and the AI on the market in 2026 is a different shape. In about six of the last ten engagements, the team running the pilot picked a vendor before there was a usable second source on the same workload, and the contract did not contemplate one. None of those teams did anything wrong at the time. The market changed underneath them in the eight months between signing and the first renewal.
No operator can run a real second-source bake-off every renewal, and that is not the move. The move is to write the next contract as if the second source will exist before the contract ends, because that is now the default, and the Apple announcement made the default visible.
SPONSORED BY CLUTCH
Hire secure AI teammates that work 24/7.
Hire pre-built AI teammates. Give your engineers and operators a platform to ship their own AI apps. Stop losing sleep about what is running where.
Clutch is the platform behind both: pre-built agents for the workflows your ops team should automate first, plus the integration plane your team's vibe-coded apps and Claude Code projects plug into. One platform. Real production. Visible and safe by default.
Built for ops, engineering, and security teams that are tired of the shadow-AI surface area inside their own company.
SINCE FRIDAY
Anthropic confidentially filed for an IPO targeting June 16 per a single-source report, which would put the first frontier-lab listing on the public market in the same window Apple's routable stack lands in developer hands. If the date holds, two of the three providers behind Siri AI will become public companies inside the same quarter.
Sriram Krishnan, the White House AI advisor, announced he is leaving at the end of the month to work on national AI competitiveness. The federal AI policy seat changes hands at the same time the EU refused Apple's 18-month phased DMA rollout and blocked Siri AI in Europe at launch. The regulatory map is now visibly different on the two continents.
SpaceX signed a $920 million per month compute agreement with Google, $11 billion per year, putting the largest commercial compute contract of the year between an X-aligned company and a Google-aligned cloud. The lab-versus-cloud lines keep moving.
Yoshua Bengio called for a coordinated pause on recursive self-improvement, arguing a verifiable agreement is the only responsible response if the frontier labs are approaching the point OpenAI flagged last week. The safety conversation is now public and has names on both sides.
Bending Spoons filed an S-1 to go public, the first AI-native rollup software company to formally test the public market this cycle. Worth watching alongside Anthropic for whether the AI-native multiple holds when the disclosure is real.
P.S. If you have an AI vendor renewal coming up in the next 90 days, hit reply and tell me what the contract says about switching costs and whether you have a second source on the same workload. I read every reply, and I am putting the patterns into next week's Memo on what enterprise AI contracts will look like after the routable-stack shift lands in procurement.
If a colleague is going into a budget review where someone is going to ask, "are we locked into the wrong vendor on AI?" forward this issue to them.
IF YOU GOT VALUE FROM THIS ISSUE
Keep going. The next 30 mornings, in your inbox.
The analysis you just read came out of the same operator practice that the 30 Days program teaches. Foundations, then connect, then build, then lead. By Day 30, you own nine concrete capabilities: prompting that hits your bar on the first draft, AI workflows that take five hours a week off your plate, agent skills your team can run, and the language to lead the AI conversation at your company.
Free. Thirty mornings, then it ends. No upsells, no community Slack. Inspired by Hilary Gridley's Couch to 5K for AI.